J. Marcos
a,
, O. Storkel
b
, L. Marroyo
a
, M. Garcia
a
, E. Lorenzo
b
a
Dpto. Ingeniera Ele ctrica y Electro nica, Universidad Pu blica de Navarra, Campus Arrosadia, 31006 Pamplona, Spain
b
Instituto de Energa Solar, EUIT Telecomunicacio n, Campus Sur UPM, 28031 Madrid, Spain
Received 30 August 2013; received in revised form 24 October 2013; accepted 28 October 2013
Communicated by: Associate Editor Elias Stefanakos
Abstract
Shortterm variability in the power generated by large gridconnected photovoltaic (PV) plants can negatively aect power quality and
the network reliability. New gridcodes require combining the PV generator with some form of energy storage technology in order to
reduce shortterm PV power uctuation. This paper proposes an eective method in order to calculate, for any PV plant size and max
imum allowable ramprate, the maximum power and the minimum energy storage requirements alike. The general validity of this method
is corroborated with extensive simulation exercises performed with real 5s one year data of 500 kW inverters at the 38.5 MW Amaraleja
(Portugal) PV plant and two other PV plants located in Navarra (Spain), at a distance of more than 660 km from Amaraleja.
2013 Elsevier Ltd. All rights reserved.
Keywords: Gridconnected PV plants; Power uctuations smoothing; Ramprate control; Energy storage sizing
1. Introduction
Concerns about the potential of PV output uctuations
caused by transient clouds were expressed more than
25 years ago (Jewell and Ramakumar, 1987; Jewell and
Unruh, 1990) and are now attracting widespread interest
and attention, as a result of growing PV penetration rates.
As the PV power share in the grid increases, such uctua
tions may adversely aect power quality and reliability
(Marcos et al., 2011a). In particular, power uctuations
of less than 10 min are typically absorbed by the grid as
frequency uctuations. This issue is of special importance
in relatively small grids, such as islands, with high penetra
tion rates, because the smoothing eect from the aggrega
tion of geographically dispersed PV plants is intrinsically
limited (Marcos et al., 2011b; Perpin an et al., 2013). It
was precisely an island grid operator, The Puerto Rico
Electric Power Authority, that recently opened the door
for PV power variability regulations, by imposing a 10%
per minute rate (based on nameplate capacity) limitation
on the PV plants being connected to its grid (PREPA,
2012).
Standard (without storage) PV plants exhibit power
variations far beyond this limitation. For example, up to
90% and 70% per minute variations have been recorded,
respectively, at 1 MW and 10 MW PV plants (Marcos
et al., 2010). Hence, compliance with such regulations
requires combining the PV generator with some form of
energy storage technology, to either add or subtract power
to or from the PV output in order to smooth out the high
frequency components of the PV power. Fuel cells
(Rahman and Tam, 1988), electricdouble layer capacitors
(Kakimoto et al., 2009) and, mainly, batteries (Hund et al.,
2010; Byrne et al., 2012; Ellis et al., 2012; Leitermann,
2012; Xiangjun et al., 2013) have been proposed. Smooth
ing algorithms can be found (Kakimoto et al., 2009; Hund
0038092X/$  see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.solener.2013.10.037
. That is:
DP
Dt
t
Pt Pt Dt
P
100 % 1
It is then possible to compare the time series of DP
Dt
(t)
with a given ramp value, r, and count the time the uctua
tions exceed the ramp (abs [DP
Dt
(t)] > r). Fig. 2 shows the
results for a full year (July 2010June 2011) and for the dif
ferent Amaraleja PV sections described above. As
expected, the occurrence of uctuations decreases with r
and with P
= 38.5 MW
Fig. 1. Field distribution of the Amaraleja PV plant sections considered in
this work.
Fig. 2. Frequency over one year (July 2010June 2011) of PV power
uctuations calculated in 1min time window, DP
1min
(t), are superior to a
given ramp r (%/min). The frequency value is given in relative terms to the
total production time (4380 h).
J. Marcos et al. / Solar Energy 99 (2014) 2835 29
reduces the time the ramp is exceeded to 23%, whilst for a
much less stringent ramp, r = 30%/min, these values drop
to 3% and 0.1%, respectively. These examples show that
imposing power ramp limits (typically around 10%/min)
makes it necessary to resort to an Energy Storage System
(ESS) even when large PV plants are concerned. In this
paper we will assume that our ESS is a battery, just to
make the presentation easier, although all the analyses
shown are equally valid for any other storage technology.
4. Smoothing of power uctuations by energy storage
4.1. Ramprate control
Let us consider a maximum permissible ramp rate value
of the power injected into the grid, r
MAX
(%/min). Fig. 3
shows a basic model of the corresponding ramprate con
trol. P
PV
(t), P
G
(t) and P
BAT
(t) are, respectively, the power
from the inverter, the power to the grid and the power to
the battery. Obviously:
P
BAT
t P
G
t P
PV
t 2
Initially, the inverter tries to inject all its power into the
grid, P
G
(t) = P
PV
(t). The control is activated when the
maximum allowable ramp condition is broken. That is, if:
jDP
G;1 min
tj > r
MAX
3
Fig. 3. Ramprate control model for a given P
PV
(t) time series. Looking
for simplicity, battery and associate electronic converter losses are
ignored.
Fig. 4. (a) Evolution of the generated power, P
PV
(t) by Section B (1.1 MW) on 31th October 2010 and the simulated power which would be injected to the
grid P
G
(t) in the case of disposing a battery which limits uctuations to r
MAX
of 10%/min (0.833%/5 s). (b) Battery power, P
BAT
. (c) Battery energy, E
BAT
.
30 J. Marcos et al. / Solar Energy 99 (2014) 2835
Then, the corresponding power excess or shortage is
either taken from (P
BAT
(t) > 0) or stored into (P
BAT
(t)< 0)
the battery. The energy stored at the battery, E
BAT
(t), is
given by the integral of P
BAT
(t) over time. In this way,
the behavior of the whole system can be easily simulated
for any time series of P
PV
(t). For the sake of simplicity,
any potential battery and associated electronic converter
losses are disregarded here.
As a representative example, Fig. 4 shows, for
r
MAX
= 10%/min, the 1.1 MW Amaraleja PV section on
an extremely uctuating day (31th October, 2010), the
resulting evolution of P
PV
(t) and P
G
(t) (Fig. 4a), P
BAT
(t)
(Fig. 4b) and E
BAT
(t) (Fig. 4c). Battery requirements for
this day derive from the corresponding maximum power
and energy values. In this example, the required battery
power is P
BAT,MAX
= 873 kW (or P
BAT,MAX
= 0.79P
)
and the required battery capacity is C
BAT
= E
BAT,MAX

E
BAT,MIN
= 175 kW h (or 10 min of capacity, equivalent
to 0.16 h of PV plant production at P
). It is worth men
tioning that the daily battery energy balance is negative
(20 kW h). At rst glance, this may appear counterintu
itive, because the PV power uctuation distribution is
essentially symmetrical (clouds reaching and leaving the
PV eld). However, this can be understood by carefully
observing the battery charge and discharge dynamic. Note
that the area of upper regions (charging) is larger than the
area of lower ones (discharging).
Fig. 5 shows the result of extending the simulation exer
cise to an entire year (July 2010June 2011), to all the
Amaraleja PV sections, and for r
MAX
= 10%/min. The
State of Charge (SOC) of the battery at the end of a day
has been concatenated with the SOC at the beginning of
the next day. As the example shown in Fig. 4, the tendency
of the battery to discharge continuously aects the entire
one year period. An important initial conclusion can be
drawn: instead of distributing the storage systems for single
power plants or sections within a power plant, it seems
wiser to add multiple sections or power plants to a single
storage system. On the other hand, the battery discharging
tendency leads to excessive battery capacity requirements,
in the order of some hours. More practical alternatives
are obtained when adding charge to the battery at dierent
times throughout the year, as will be seen below. Neverthe
less, an important conclusion can be reached from Fig. 5:
the energy that must be managed through the storage sys
tems is very low, only about 0.3% of the total energy pro
duction for limiting the power ramps of a 0.5 MW plant at
a maximum of 10%/min (for this case, as Fig. 2 showed, the
battery time of use is equal to 8%). Thus, eciency related
aspects are scarcely relevant.
4.2. Overnight battery recharging
Overnight battery recharging from the grid makes sense
because electricity demand usually drops at night. Fig. 6
presents the results of a simulation exercise similar to
Fig. 5, except that this time, if required, energy at the bat
tery is restored each night. That is:
E
BAT ;end day
i1
< 0 ) E
BAT;beginning day
i
0 4
In this way the tendency of the battery to discharge con
tinuously does not aect the entire period of one year, but
is limited to one day and therefore signicantly reduces the
required battery size, which is now in the order of some
minutes. For example, battery requirements for
r
MAX
= 10%/min in the 1.1 MW Amaraleja PV section
are now P
BAT,MAX
= 890 kW (or P
BAT,MAX
= 0.81P
)
and E
BAT,MAX
= 451 kW h (or 25 min of capacity, equiva
lent to 0.41 h of PV plant production at P
). The compar
ison of these gures with the above mentioned results for
31th October 2010, reveals that power battery require
ments, which are obviously imposed by the worst individ
ual uctuation, tend to be constant throughout the
analysis period. However, the same is not true for the bat
tery energy requirements, which are imposed by the uctu
ation distribution throughout the worst day.
4.3. Daytime battery recharging controlled by state of charge
Another interesting battery recharging possibility, not
requiring energy to be supplied from the grid, consists in
establishing a reference value for the energy stored in the
Fig. 5. Evolution of storage time, E
BAT
/P
) and
C
BAT
= E
BAT,MAX
E
BAT,MIN
= 124 kW h (or 6.7 min of
capacity, equivalent to 0.11 h of PV plant production at
P
to 0.1P
to P
)
with a time constant, s (s), which is empirically correlated
(Fig. 10) with the shortest dimension of the perimeter of
the PV plant, l (m), by an expression such as:
s a l b 5
where a = 0.042 (s/m) and b = 0.5 s. Table 1 presents the
real s values observed at the dierent PV Amaraleja sec
tions and Fig. 10 shows that they are in good agreement
with Eq. (5).
Fig. 7. Ramprate control model modied with additional SOC control.
Notice that the SOC control action is also smoothed by the ramplimiter
in order to guarantee that power uctuations are always below r
MAX
.
Fig. 8. Evolution along July 2010June 2011 of E
BAT
(a) and P
BAT
(b) for section B (1.1 MW), a r
MAX
of 10%/min and with SOC control. E
BAT,REF
and K
have been arbitrarily set to 175 kW h and 6, respectively.
Fig. 9. Worst uctuation model. The blue line represents the P
PV
(t)
response to an irradiance uctuation (yellow line) and the red one is the
power injected to the grid P
G
with a ramprate control. The dierence
between P
G
and P
PV
is P
BAT
, the maximum dierence corresponds to
P
BAT,MAX
and the dened integral of P
BAT
corresponds to E
BAT,MAX
. (For
interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)
32 J. Marcos et al. / Solar Energy 99 (2014) 2835
Battery requirements for ramprate limitation are easily
derived from the model showed in Fig. 9. We can see the
response of P
PV
(t) and P
G
(t) to a negative irradiance G(t)
uctuation. P
PV
(t) evolution corresponds to a rst order
system with a time constant s, while P
G
decreases with a
rhythm being set by r
MAX
. The power demanded to the bat
tery P
BAT
(t) corresponds with the dierence between P
G
(t)
and P
PV
(t), Eq. (2). Therefore, P
BAT
(t) along the worst
uctuation time is given by:
P
BAT
t
P
100
901 expt=s t r
MAX
6
where r
MAX
is expressed as % per time. This expression gets
a maximum for
t
P
BAT;MAX
s ln
90
s r
MAX
7
Thus, the required battery power is given by:
P
BAT;MAX
t
P
100
90 s r
MAX
1 ln
90
s r
MAX
8
where P
, P
BAT,MAX
is expressed in (kW), r
MAX
in (%/s)
and s in (s). On the other hand, the battery discharging
process lasts until the time the power ramp reaches 0.1P
.
Corresponding time span, T
R
, is:
T
R
90
r
MAX
9
Thus, the required battery energy is given by:
E
BAT ;MAX
Z
T
R
0
P
BAT
tdt
0:9P
3600
90
2 r
MAX
s 1 exp
90
s r
MAX
0:9P
3600
90
2 r
MAX
s
10
where P
is expressed in (kW), r
MAX
in (%/s), s in (s) and
E
BAT,MAX
in (kW h). As the sign of the rst uctuation is
unknown, a double capacity battery is required to absorb
both the upwards and downwards uctuation:
Fig. 10. Adjustment of observed time constant values s vs. shortest
perimeter dimension l, Eq. (5). The general expression of this equation is
y = mx + n, where m gives the coherency to the units. In our case,
m = 0.042 (s/m).
Table 1
Characteristic power P
, in hours.
Results derived from the worst uctuation model show good agreement with the ones derived from detailed simulation based on 5 s real data recorded at
dierent Amaraleja PV sections.
J. Marcos et al. / Solar Energy 99 (2014) 2835 33
C
BAT
2 E
BAT;MAX
1:8P
3600
90
2 r
MAX
s
11
For example, for P
= 38.5 MW and
l = 1786 m, corresponding results are s = 74.51s,
P
BAT,MAX
= 0.53P
= 20.4 MW and C
BAT
= P
0.098 h =
3773 kW h.
Fig. 11 compares the battery requirements for the dier
ent PV Amaraleja sections and for dierent ramprate
limits, as deduced from simulation based on a year of
observed 5 s data and as given by Eqs. (8) and (11). Good
agreement is clearly observed. Furthermore, in order to
check the general validity of the worst uctuation model,
we performed a similar exercise for two dierent PV plants
located at a distance of about 660 km from Amaraleja, at
Rada (P